The responses of measurement systems are often observed as curves. In this paper, a singular-value decomposition technique is applied to the following three problems in the analysis of response curves: (1) extraction of components or features, (2) effective representation of curves, (3) estimation of unknown curves. The analysis is based on that a response curve can be decomposed to become a linear combination of basic-component patterns, and the predominant members of these patterns provide a good approximate representation of the original response curve. The complicated estimation problem of interpolating the curves, then, can be reduced to a simple interpolation of weighting coefficients needed to combine the component patterns.
Next, the proposed method is applied to the measured curves of γ-ray response spectra. The results show that the curves of response spectra can be represented by utilizing only two principal-component patterns. Moreover, we discuss about the accuracy of the estimated spectral curves and the improved method, and illustrate the goodness of the interpolated results.